Fig. 3: Guided Backpropagation activation maps of the tested surfaces. | Communications Engineering

Fig. 3: Guided Backpropagation activation maps of the tested surfaces.

From: A deep learning framework for predicting the effect of surface topography on thermal contact resistance

Fig. 3: Guided Backpropagation activation maps of the tested surfaces.

a and g show the activation maps of the original ground and turned surfaces, respectively. b and h depict the activation maps of the ground and turned surfaces rotated by 90°. The predicted actual contact area varies with the rotation angle at a contact pressure (P) of 1 MPa for the ground (c) and turned (i) surfaces. Similarly, the predicted thermal contact resistance (TCR) changes with the rotation angle at 1 MPa for the ground (d) and turned (j) surfaces. The predicted actual contact area fluctuates with the sum of gradients for the ground (e) and turned (k) surfaces at 1 MPa. Finally, the predicted TCR varies with the sum of gradients for the ground (f) and turned (l) surfaces at 1 MPa. Here, Surf1 and Surf2 denote height profiles of the two contact surfaces, while Surf1 + Surf2 represents the direct summation of the height tensors of these surfaces. Similarly, Grad1 and Grad2, as well as Grad1 + Grad2, correspond to the dimensions of the Guided Backpropagation gradient matrices associated with the respective input surfaces.

Back to article page